(a) Field of the Invention:
The present invention relates to a control system for a disk cache memory used between a main memory and a disk unit.
(b) Related Art Statement:
For improvement of the effective processing speed of a disk unit, a disk cache memory system has been proposed in which a cache memory is used between the main memory unit and the disk unit. With an input/output instruction from the central processing unit, the disk cache system will access the cache memory if object data is found in the latter. If not, simultaneously with, before or after the input/output process, the data near the object data are loaded in a certain amount from the disk unit into the cache memory to meet the data request of an input/output instruction which will be issued later. This data loading is done per record, track, several tracks, cylinder, data set or the like. Generally, however, in case a variable-length record is handled, the data loading is done per track or several tracks in many cases. Since it is evident that particularly in a process of reading a sequence of data sets in which access to a record is sequentially done, data is read sequentially from one record after another, it is effective to read data from several tracks and load them into the cache memory. A technique related to such a system is known from the Japanese Unexamined Patent Publication No. 55-157056.
In case the data handled with one input/output instruction is long, it can be effectively loaded using a long unit but not the sequential access.
However, no consideration was given to the fact that loading such long data all at once into the cache memory will take a longer time. Namely, in this conventional technique, an input/output instruction is issued during a long time of data loading, and if it is found in the cache memory (this is called "miss" and existence of such instruction in the cache memory is called "hit"), access to the disk memory is put off until the data loading has been completed, which results in an increased input/output processing time. This wait time is dozens of milliseconds even for the data loading from one track. Loading from several tracks or cylinders will take several times or dozens of times as long time as the wait time for data loading from one track. Therefore, in case the probability of "hit" is low, data loading will increase causing a very large overhead.